Summary The “selfish herd” hypothesis1 provides a potential mechanism to explain a ubiquitous phenomenon in nature: that of non-kin aggregations. Individuals in selfish herds are thought to benefit by reducing their own risk at the expense of conspecifics by attracting toward their neighbors’ positions1,2 or central locations in the aggregation.3, 4, 5 Alternatively, increased alignment with their neighbors’ orientation could reduce the chance of predation through information sharing6, 7, 8 or collective escape.6 Using both small and large flocks of homing pigeons (Columba livia; n = 8–10 or n = 27–34 individuals) tagged with 5-Hz GPS loggers and a GPS-tagged, remote-controlled model peregrine falcon (Falco peregrinus), we tested whether individuals increase their use of attraction over alignment when under perceived threat. We conducted n = 27 flights in treatment conditions, chased by the robotic “predator,” and n = 16 flights in control conditions (not chased). Despite responding strongly to the RobotFalcon—by turning away from its flight direction—individuals in treatment flocks demonstrated no increased attraction compared with control flocks, and this result held across both flock sizes. We suggest that mutualistic alignment is more advantageous than selfish attraction in groups with a high coincidence of individual and collective interests (adaptive hypothesis). However, we also explore alternative explanations, such as high cognitive demand under threat and collision avoidance (mechanistic hypotheses). We conclude that selfish herd may not be an appropriate paradigm for understanding the function of highly synchronous collective motion, as observed in bird flocks and perhaps also fish shoals and highly aligned mammal aggregations, such as moving herds.